Tracking AI search visibility means measuring how often your brand and pages appear, are cited and are described accurately in AI answers from Google, Bing, ChatGPT, Perplexity and Gemini. Use first-party reports where they exist, such as Google’s generative AI report and Bing’s AI Performance report, plus referral analytics and a repeatable prompt test.
For years, search measurement meant rankings, impressions and clicks. AI answers broke that model: a buyer can read a summary that cites you, never click, and still contact you later. Others never see you at all because an assistant cited a competitor. This guide sets out what can actually be measured today, which tools give first-party data, how to run a manual prompt study without fooling yourself, and how to connect all of it to enquiries. It is written for marketers who need a credible report for leadership, with honest limits.
Key takeaways
- Start with first-party data. Google says its Search Console generative AI performance report was available to all sites by 31 August 2026; Bing added an AI Performance report in Webmaster Tools in February 2026.
- Clicks understate influence. Pew Research Center found users clicked links inside AI summaries on just 1% of visits in its March 2025 data.
- Prompt testing is a sample, not a ranking. AI answers vary between runs, so test a fixed question set repeatedly and report ranges.
- Track accuracy, not just presence. Being described wrongly can do more harm than not being mentioned.
- Tie it to the pipeline. Report AI visibility alongside enquiries and source data in your CRM.
What can actually be measured today?
Different surfaces offer very different levels of data. Be clear in reports about which numbers are measured and which are sampled.
| Surface | Data source | What you get | Data type |
|---|---|---|---|
| Google AI Overviews and AI Mode | Search Console generative AI performance report | Impressions for pages shown in AI features | First-party, measured |
| Google Search overall | Search Console Performance (Web) | Clicks, impressions, CTR, position; AI features included in totals | First-party, measured |
| Bing, Copilot and partners | Bing Webmaster Tools AI Performance (public preview) | Citations, cited pages, sampled grounding queries, trends | First-party, partly sampled |
| ChatGPT, Perplexity, Gemini apps | Website analytics referrals | Visits from AI assistant domains | Measured, incomplete |
| All assistants | Manual or tool-based prompt testing | Mentions, citations, accuracy, competitors | Sampled |
| Pipeline impact | CRM source fields and “how did you hear about us” | Enquiries and deals influenced | Measured plus self-reported |
Google’s generative AI performance report
Google says its generative AI performance report in Search Console shows impressions from AI Overviews and AI Mode, and that it was rolled out to all websites worldwide by 31 August 2026. Separately, Google’s AI features documentation states that traffic from AI features is included in the overall Web search type in the Performance report. Use the dedicated report to see which pages appear in AI features, and the Performance report for the full picture.
Bing’s AI Performance report
Search Engine Journal reported in February 2026 that Bing Webmaster Tools added an AI Performance report in public preview. It covers AI experiences across Copilot, AI summaries in Bing and select partner integrations, and shows total citations, average cited pages, the pages cited and sampled “grounding queries” used to retrieve content. Microsoft describes the grounding query data as a sample rather than complete activity. The report focuses on citations, not clicks. In June 2026, Bing added intents, topics, citation share and period comparison to the report; Bing describes citation share as an observational metric, not a ranking.
Why do clicks understate AI visibility?
Pew Research Center’s July 2025 study of 68,879 Google searches by 900 US adults in March 2025 found that people clicked a link inside an AI summary on just 1% of visits to pages with such a summary. They clicked a traditional result on 8% of those visits, compared with 15% when no summary appeared.
Most visits with an AI summary end without a click on its sources
A buyer may read a summary that names you, remember the name and search for you directly a week later. That journey shows up as branded search or direct traffic, not as an AI click. This is why AI visibility needs its own measures, and why branded search trends belong in the same report.
The Bargaon AI visibility scorecard
We report AI visibility as a five-layer scorecard. Each layer answers a different question, and none of them alone tells the full story.
From presence to pipeline
Report each layer separately; do not merge them into one “AI score”.
| Layer | Metric | Source | Frequency |
|---|---|---|---|
| Access | Key pages crawlable by Googlebot, Bingbot, OAI-SearchBot, PerplexityBot | robots.txt review, server logs, URL Inspection | Quarterly and after site changes |
| Presence | AI feature impressions by page | Search Console generative AI report | Monthly |
| Presence | Citations and cited pages | Bing Webmaster Tools AI Performance | Monthly |
| Share | Share of answers mentioning the brand, by question group | Prompt study | Quarterly |
| Accuracy | Answers with a factual error about the brand | Prompt study | Quarterly |
| Impact | Branded impressions and clicks | Search Console query filter | Monthly |
| Impact | Visits from AI assistant domains | Analytics referral report | Monthly |
| Impact | Enquiries mentioning AI or with AI referral source | CRM | Quarterly |
How do you run a prompt study that means something?
Prompt testing is the only way to see answers from assistants that offer no first-party reports. Done carelessly, it produces impressive-looking but meaningless numbers. These rules keep it honest:
- Fix the question set. Choose 20 to 50 real buyer questions grouped by stage (problem, solution, comparison, provider). Keep the wording identical every time.
- Fix the conditions. Use the same assistants, logged out or in clean sessions, with no personal memory, from the same country.
- Repeat each question. Run each one at least three times per assistant per round, because answers vary between runs.
- Record the same fields. Brand mentioned (yes/no), site cited (yes/no), position of mention, competitors named, factual errors, and the cited URLs.
- Report ranges, not single numbers. “Mentioned in 3 to 5 of 10 comparison questions” is more honest than “40% visibility”.
- Keep the raw log. Store answers or screenshots so others can check the scoring.
Example prompt study log
| Question | Assistant | Run | Brand mentioned | Site cited | Competitors named | Error? |
|---|---|---|---|---|---|---|
| Best technical SEO consultants for B2B in India | Perplexity | 1 | Yes | Yes | 3 | No |
| Best technical SEO consultants for B2B in India | Perplexity | 2 | No | No | 4 | — |
| How to plan a website migration without losing traffic | ChatGPT | 1 | No | Yes | 0 | No |
| HubSpot vs Zoho for a small B2B team | Gemini | 1 | No | No | — | — |
The rows are illustrative, to show the format. They are not real results.
How do you track AI referral traffic?
Some assistants pass a referrer when a user clicks a cited link. In your analytics tool, create a channel group or segment that matches known assistant domains, such as chatgpt.com, perplexity.ai, gemini.google.com and copilot.microsoft.com. Review the list regularly, because domains change. Note the limits:
- Some apps strip the referrer, so visits appear as direct traffic.
- Google AI Overviews and AI Mode clicks are reported as Google organic search, not as a separate referrer.
- Volumes are often small. Look at the quality of these visits, such as engagement and enquiries, not only the count.
Our guide to marketing analytics and attribution explains how to carry source data into your CRM so enquiries can be traced.
Should you use a paid AI visibility tool?
A growing number of tools automate prompt testing across assistants. They can save time on large question sets, but they share the same limits as manual testing: answers vary, tools may test from locations or settings that differ from your buyers, and their “visibility scores” are proprietary calculations. If you use one, ask:
- Which assistants and models are tested, from which country, and how often?
- How many runs per question, and how is variation handled?
- Can you export raw answers and citations?
- How is the score calculated, and can you reproduce it?
For most small and mid-size businesses, Search Console, Bing Webmaster Tools, analytics and a quarterly manual study of 30 questions are enough to make good decisions.
How should you report AI visibility to leadership?
Keep the report short and honest. A one-page quarterly view works well:
- Headline: what changed, in one or two sentences.
- Presence: AI feature impressions and Bing citations, with the trend.
- Share and accuracy: prompt study results as ranges, with notable errors and fixes.
- Impact: branded search trend, AI referrals and enquiries that mention AI.
- Actions: the three to five pages or sources to improve next quarter.
Avoid combining these into a single “AI visibility score”. It hides the detail leadership needs and invites false precision.
How do you act on what you find?
Measurement is only useful if it changes what you do. Typical actions by finding:
- Not crawlable: fix robots.txt and rendering; see how ChatGPT, Perplexity and Gemini choose sources.
- Crawlable but not cited: improve answers, depth and evidence; see Google AI Overviews and answer engine optimization.
- Competitors cited instead: compare their cited pages with yours and close the gaps; build third-party presence.
- Described wrongly: correct the source pages and profiles; see entity SEO and structured data.
How do you set up AI visibility tracking in the first 30 days?
A baseline is more useful than any single report. In the first month, set up the following in order:
- Week 1, access check. Review robots.txt for Googlebot, Bingbot, OAI-SearchBot and PerplexityBot. Confirm key pages return 200, are indexable and render their main text in HTML. Note any blocks that are deliberate.
- Week 1, first-party reports. Verify the site in Search Console and Bing Webmaster Tools if not already done, submit the XML sitemap to both, and note the date the generative AI and AI Performance reports first show data.
- Week 2, analytics segment. Create an “AI assistants” channel or segment using known referrer domains. Add a landing-page breakdown so you can see which pages AI visitors arrive on.
- Week 2, CRM fields. Make sure enquiry forms capture source and landing page automatically, and add an optional “How did you hear about us?” field with an “AI assistant (ChatGPT, Perplexity, etc.)” option.
- Week 3, question set. Agree 20 to 50 buyer questions with sales, grouped by stage. Freeze the wording.
- Week 4, first prompt study. Run the baseline study and store the raw answers.
By the end of the month you have a baseline for every layer of the scorecard. Every later report compares against it.
How should the prompt question set be built?
The quality of a prompt study depends on the questions. A good set reflects how real buyers ask, not how marketers describe the offer. Build it in four groups, with five to twelve questions in each:
| Group | Buyer’s situation | Example question | What a good result looks like |
|---|---|---|---|
| Problem | Knows something is wrong, not the solution | “Why are our website enquiries dropping?” | Your guide is cited as a source |
| Solution | Knows the type of solution | “How do I set up lead tracking from website to CRM?” | Your method or framework is cited |
| Comparison | Choosing between options | “HubSpot or Zoho for a 10-person B2B team?” | Your comparison is cited and accurate |
| Provider | Choosing who to hire | “Who can help set up Zoho CRM for a small business in India?” | Your brand is named and described correctly |
Take the wording from sales calls, enquiry forms and Search Console queries rather than inventing it. Keep a copy of the frozen list with the date it was agreed, so later studies use exactly the same questions.
How should you interpret changes over time?
| What you see | Likely meaning | Next step |
|---|---|---|
| AI feature impressions up, clicks flat | More visibility in summaries, fewer click-throughs | Check branded search and enquiries before judging |
| Bing citations up for a few pages only | Those pages answer grounding queries well | Apply the same structure to related pages |
| Prompt study mentions fall | Competitors improved, or answers changed with a model update | Compare cited sources; repeat next month to confirm |
| Repeated factual error | Outdated or conflicting source somewhere on the web | Find and fix the source; re-test in a quarter |
| AI referrals rise but enquiries do not | Visitors land on informational pages without a next step | Add relevant service links and calls to action |
Look for patterns across at least two reporting periods before acting on a single change. AI surfaces are noisy, and one month rarely tells the full story.
Who should own AI visibility reporting?
Give one person ownership of the scorecard, usually the SEO or content lead, with support from analytics for referral and CRM data and from sales for the question set. The owner runs the monthly report, the quarterly prompt study and the review meeting, and keeps the raw logs. Without a named owner, AI visibility tends to be checked occasionally by different people using different questions, which produces numbers that cannot be compared. A short monthly check of 30 minutes and a quarterly review of two to three hours is enough for most small and mid-size businesses.
What AI visibility tracking cannot tell you
No method shows every AI answer that mentions you. First-party reports cover only their own platforms, and Bing’s grounding queries are sampled. Prompt studies are small samples of a moving target. Attribution from an AI answer to a deal is often impossible to prove. Use these measures to spot direction and problems, not to claim precise market share. For the strategy these measures support, see our generative engine optimization guide.
Frequently asked questions
Can I see which AI Overviews cite my page?
Search Console’s generative AI performance report shows impressions for pages appearing in AI Overviews and AI Mode. Check Google’s help page for the exact dimensions available in your account.
Does ChatGPT offer a reporting tool for publishers?
Not at the time of writing. Use analytics referrals from chatgpt.com and a prompt study.
How many questions should a prompt study include?
Twenty to fifty real buyer questions, grouped by stage, is a practical range for most businesses. Run each several times per assistant.
Why do my prompt study results change every time?
AI answers vary between runs and change as models and indexes update. That is why you repeat questions and report ranges.
Is branded search a good proxy for AI visibility?
It is a useful supporting signal. If AI answers mention you more, some buyers will search for your name. Watch branded impressions alongside first-party AI reports.
Sources and further reading
- Search Console Help: Generative AI performance report.
- Google Search Central: AI features and your website.
- Search Engine Journal: Bing Webmaster Tools adds AI citation performance data (February 2026).
- Bing Search Blog: New AI visibility insights in Bing Webmaster Tools (June 2026).
- Pew Research Center: AI summaries and clicks (July 2025).
- OpenAI: crawler documentation.
Need a credible AI visibility report? Bargaon’s generative engine optimization service sets up the scorecard, runs the first prompt study and turns the findings into a page-by-page action plan.